US20030071669A1

US20030071669A1 - Deskewing global clock skew using localized DLLs

Info

Publication number: US20030071669A1
Application number: US09/975,359
Authority: US
Inventors: Dean Liu; Tyler Thorp; Pradeep Trivedi; Gin Yee; Claude Gauthier
Original assignee: Sun Microsystems Inc
Current assignee: Oracle America Inc
Priority date: 2001-10-11
Filing date: 2001-10-11
Publication date: 2003-04-17
Anticipated expiration: 2021-10-11
Also published as: GB0408304D0; US6686785B2; WO2003032137A2; WO2003032137A3; GB2399468A

Abstract

A method for reducing global clock skew by referencing a first point on an integrated circuit to which to align other points on the integrated circuit is provided. Further, an integrated circuit that has localized DLLs having adjustable buffers that selectively drive a signal on a clock grid is provided. Further, a technique for using a local DLL, one or more phase detectors, and one or more DLLs connected to portions of a clock grid to reduce clock skew is provided.

Description

BACKGROUND OF INVENTION

A typical computer system includes at least a microprocessor and some form of memory. The microprocessor has, among other components, arithmetic, logic, and control circuitry that interpret and execute instructions necessary for the operation and use of the computer system. FIG. 1 shows a typical computer system ( 10) having a microprocessor (12), memory (14), integrated circuits (16) that have various functionalities, and communication paths (18), i.e., buses and wires, that are necessary for the transfer of data among the aforementioned components of the computer system (10).

The components of a computer system use a reference of time to perform the various operations of the computer system. This reference of time is provided to the components of the computer system using one or more clock signals. The components use the one or more clock signals to determine when to conduct certain operations. As computer systems continue to operate at ever-increasing frequencies, it becomes more and more important to ensure that the components of a computer system receive their clock signals in an accurate and timely manner because a mistiming has the potential to cause an error, performance setback, or an outright malfunction of the computer system.

FIG. 2 shows a clock distribution network ( 20) for a microprocessor (12). A reference clock (also known in the art as “system clock” and shown in FIG. 2 as REF_CLK), which is typically generated from outside the microprocessor (12), serves as an input to a phase locked loop (“PLL”) (15). Essentially, the PLL (15) uses feedback to maintain a specific phase relationship between its output (shown in FIG. 2 as CHIP_CLK) and the reference signal. The chip clock from the PLL (15) is then distributed to one or more clock drivers/buffers (17), which, in turn, distribute the chip clock to a global clock grid (19), where the global clock grid (19) feeds the chip clock to various microprocessor components such as local clock grids (24) and a feedback loop (26) that feeds the chip clock back to the PLL (14). The local clock grids (24) feed the chip clock to base components of the microprocessor (12), such as latches (22) and flip-flops (28).

As a clock signal, such as the chip clock shown in FIG. 2, is propagated to the various parts and components of a microprocessor, one or more types of system variations may alter the behavior and/or integrity of the clock signal. Common system variations include, but are not limited to, power variations, temperature variations, and process variations. Due to these and other variations across a microprocessor, a particular clock signal may arrive at different parts of the microprocessor at different times. This difference in the arrival of a clock signal at different system components is referred to and known in the art as “clock skew.”

As partly discussed above, clock skew is a function of architectural factors such as load, device distribution across a microprocessor, device mismatch, and temperature and voltage gradients across the microprocessor. By designing a microprocessor that accounts for some of these variations, the amount of clock skew in the microprocessor may be reduced. The process of removing or decreasing clock skew is referred to and known in the art as “deskewing.”

Clock deskewing is typically performed in an upper distribution layer of a clock distribution network. For example, in a clock distribution network that has a global and a local layer, deskewing is performed in the global layer. Similarly, in a network that has a global, a regional, and a local layer, deskewing is performed in the global and/or regional layers.

FIG. 3 shows a typical clock distribution network ( 40) having a global distribution layer (42), a regional distribution layer (44), a local distribution layer (46), where clock deskewing occurs in the regional distribution layer (44). In FIG. 3, a PLL (48) distributes a chip clock to a set of one or more clock drivers/buffers (50), which, in turn output the chip clock to a set of deskewing buffers (52) in the regional distribution layer (44). The deskewing buffers (52) deskew the chip clock and then distribute the deskewed chip clock to a global clock grid (54), which is connected to one or more local clock grids (58), where the local clock grids (58) are connected to microprocessor components such as latches (56), flip-flops (60), and other types of circuit elements (not shown).

A deskewing buffers, as shown in FIG. 3 are typically implemented as a delay locked loop (DLL). A DLL is a component that uses a control signal to maintain an output signal in a specific delay relationship with an input signal. The control signal indicates to the DLL how much delay, if any, the DLL needs to insert into the output signal. Because the amount of delay a DLL inserts is typically not a constant or predefined value, the DLL is known as a “variable delay circuit.”

As shown in FIG. 3, when deskewing buffers are included in the regional distribution layer of a clock distribution network, adjusting global clock skew is a less onerous task. However, such deskewing does not account for clock skew contributed by devices and variations in the local distribution layer (such clock skew is referred to and known in the art as “localized clock skew”).

SUMMARY OF INVENTION

According to one aspect of the present invention, an integrated circuit comprises a local delay locked loop, an adjustable delay locked loop comprising a tunable buffer that is connected to a clock grid, and a phase detector that indicates to the adjustable delay locked loop whether the tunable buffer needs to be modulated based on a reference clock and a feedback clock, where the reference clock is operatively connected to the local delay locked loop, and where the feedback clock is operatively connected to the clock grid.

According to another aspect, a method for reducing clock skew on a clock grid comprises inputting a reference clock operatively connected to a local delay locked loop, where the local delay locked loop resides at a first location, inputting a feedback clock operatively connected to the clock grid, determining whether the feedback clock needs to be modulated based on the reference clock, and selectively adjusting a delay of a buffer connected to the clock grid dependent on the determination.

According to another aspect, a method for decreasing clock skew comprises referencing a point on a clock grid to which to align other points on the clock grid by varying at least one delay of at least one tunable buffer connected to the clock grid.

According to another aspect, an integrated circuit having a clock grid comprises referencing means for referencing a first point on the clock grid to align at least one other point.

Other aspects and advantages of the invention will be apparent from the following description and the appended claims.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 shows a typical computer system. [0015]
FIG. 2 shows a typical clock distribution network. [0016]
FIG. 3 shows a typical clock distribution network. [0017]
FIG. 4[0018] a shows a component layout in accordance with an embodiment of the present invention.
FIG. 4[0019] b shows a component layout in accordance with the embodiment shown in FIG. 4a.
FIG. 5 shows a design in accordance with an embodiment of the present invention.[0020]

DETAILED DESCRIPTION

Embodiments of the present invention relate to a method for reducing global clock skew in an integrated circuit. Embodiments of the present invention further relate to a circuit device implementation that reduces global clock skew. Embodiments of the present invention further relate to a technique for increasing computer system performance by reducing system uncertainties associated with clock skew in the computer system. [0021]
Particularly, the present invention relates a technique for referencing a point on a clock grid for which to align other points on the clock grid by varying one or more delays of clock buffers associated with the clock grid. This technique helps decrease clock skew by reducing systematic components of clock skew in an integrated circuit. [0022]
FIG. 4[0023] a shows an exemplary component layout of an integrated circuit (80) in accordance with an embodiment of the present invention. A local DLL (82) of the integrated circuit (80) outputs to one or more adjustable DLLs (84) (shown in FIG. 4a as DLL′). This output from the local DLL (82) serves a feedback clock to a PLL (not shown). The adjustable DLLs (84) use the output of the local DLL (82) as a reference to lock to. Specifically with reference to FIG. 4a, the adjustable DLLs (84) in the top and bottom rows lock to the adjustable DLLs (84) in the middle row and the adjustable DLLs (84) in the middle row lock to the local DLL (82). Positioned in between a pair of adjustable DLLs (84) is a phase detector (86), where the phase detector (86) outputs to an associated adjustable DLL (84). Delay elements within the adjustable DLLs (84) are replicated within a region of a global clock grid (not shown). Those skilled in the art will appreciate that such replication may result in replacing a portion of a global clock grid driver.
FIG. 4[0024] b shows a detailed structure of a section (87) of the integrated circuit (80) in accordance with the embodiment shown in FIG. 4b. The phase detector (86) receives both a reference clock (shown in FIG. 4b as REF_CLK) from an adjacent adjustable DLL (not shown in FIG. 4b) and a feedback clock (shown in FIG. 4b as FEEDBACK) from the adjustable DLL (84) to which it is associated. The phase detector (86) determines whether a time phase of the feedback clock is aligned with the time phase of the reference clock. When the time phases of the feedback clock and the reference clock are not aligned, the phase detector (86) indicates as such on an up/down signal (shown in FIG. 4b as UP/DOWN) to a finite state machine (88) (shown in FIG. 4b as FSM) inside the adjustable DLL (84). The up/down signal generated by the phase detector (86) indicates whether the feedback clock needs to be sped up (by an “up” indication) or whether the feedback clock needs to be slowed down (by a “down” indication). Those skilled in the art will appreciate that the reference clock and feedback clock into the phase detector (86) may be implemented in one or metal layers of an integrated circuit in order to decrease the effect of transistor and voltage variations.
The finite state machine ([0025] 88) uses a counter function that counts the number of times the up/down signal is up and the number of times the up/down signal is down. Using these counts, the finite state machine (88) generates control bits to a multi-bit control bus (89) that is used to module the delay of one or more tunable buffers (90).
The tunable buffers ([0026] 90) in the adjustable DLL (84) are essentially delay elements that are replicated and distributed across a global clock grid (94). The adjustable buffers (90) are interleaved between regular global clock grid buffers (92). With a proper control setting into the tunable buffers (90), the delay of the global clock grid (94) may be modulated. For example, as a default control setting, a tunable buffer (90) has the same drive strength as a regular global clock grid buffer (92). As the control changes, the resulting delay of the tunable buffers (90) is interpolated with that of the regular global clock grid buffers (92) over the global clock grid (94), thereby changing an overall delay of the global clock grid (94).
FIG. 5 shows an exemplary design of a tunable buffer ([0027] 90) in accordance with an embodiment of the present invention. The tunable buffer (90) is implemented using parallel transistor stacks (94). The gate of each transistor in a transistor stack in is connected to a particular control bit on the control bus (89). When a control bit is asserted, the corresponding transistor stack contributes to an output current of the tunable buffer (90), thereby increasing the tunable buffer's drive strength. Conversely, when the control bit is deasserted, the corresponding transistor stack is tri-stated, thereby reducing the tunable buffer's drive strength.
Those skilled in the art will appreciate that the number of control bits used to control a particular tunable buffer may be adjusted depending on the amount of skew that one wishes to adjust for and/or the amount of delay resolution one wishes to observe or have. It follows that using more control bits allows for a larger range of compensation and finer time step. Further, those skilled in the art will appreciate that the number of transistor stacks used in a tunable buffer may also be changed according to the amount of control bits one wishes to use. [0028]
Advantages of the present invention may include one or more of the following. In some embodiments, because a point on a clock grid is referenced to align other points on the clock grid to, clock skew is reduced. [0029]
In some embodiments, because localized adjustable DLLs are used to reduce global clock skew, clock skew introduced at the local distribution layer may be reduced. [0030]
In some embodiments, because clock skew is decreased by using localized DLLs having tunable buffers that are connected to a global clock grid, a skew budget for a clock distribution network may be decreased. [0031]
In some embodiments, because clock skew is reduced, system performance is increased. [0032]
While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this disclosure, will appreciate that other embodiments can be devised which do not depart from the scope of the invention as disclosed herein. Accordingly, the scope of the invention should be limited only by the attached claims. [0033]

Claims

What is claimed is:

1. An integrated circuit, comprising:

a local delay locked loop;

an adjustable delay locked loop comprising a tunable buffer that is connected to a clock grid; and

a phase detector that indicates to the adjustable delay locked loop whether the tunable buffer needs to be modulated based on a reference clock and a feedback clock, wherein the reference clock is operatively connected to the local delay locked loop, and wherein the feedback clock is operatively connected to the clock grid.

2. The integrated circuit of claim 1, wherein the local delay locked loop outputs to a phase locked loop.

3. The integrated circuit of claim 1, wherein the tunable buffer is interleaved between a regular clock grid buffer and at least one other regular clock grid buffer.

4. The integrated circuit of claim 1, wherein the adjustable delay locked loop is disposed at a first location on the integrated circuit different from a second location where the local delayed lock is disposed, and wherein the adjustable delay locked loop aligns a signal at the first location with a signal at the second location by varying a delay of the tunable buffer.

5. The integrated circuit of claim 1, wherein the adjustable delay locked loop outputs the reference clock to another adjustable delay locked loop.

6. The integrated circuit of claim 1, the adjustable delay locked loop comprises a finite state machine that inputs a signal from the phase detector and outputs at least one control bit to the tunable buffer dependent on the signal from the phase detector.

7. The integrated circuit of claim 6, wherein an output from the finite state machine is a control bus.

8. The integrated circuit of claim 7, wherein the tunable buffer comprises:

a transistor stack, wherein the transistor stack contributes current to an output of the tunable buffer dependent on the control bus.

9. The integrated circuit of claim 7, wherein the transistor stack decreases current to an output of the tunable buffer dependent on the control bus.

10. The integrated circuit of claim 8, wherein the transistor stack comprises a transistor having an input operatively connected to a control bit on the control bus.

11. A method for reducing clock skew on a clock grid, comprising:

inputting a reference clock operatively connected to a local delay locked loop, wherein the local delay locked loop resides at a first location;

inputting a feedback clock operatively connected to the clock grid;

determining whether the feedback clock needs to be modulated based on the reference clock; and

selectively adjusting a delay of a buffer connected to the clock grid dependent on the determination.

12. The method of claim 11, further comprising:

generating pulses on an up/down signal to a finite state machine dependent on the determination;

generating at least one control bit to the buffer in response to generating pulses on the up/down signal; and

driving a signal on the clock grid based on the at least one control bit.

13. The method of claim 11, further comprising:

generating the reference clock to an adjustable delay locked loop.

14. An integrated circuit having a clock grid, comprising:

referencing means for referencing a first point on the clock grid to align at least one other point.

15. The integrated circuit of claim 14, the referencing means comprising:

adjusting means for varying a driver connected to the clock grid.

16. A method for decreasing clock skew, comprising:

referencing a point on a clock grid to which to align other points on the clock grid by varying at least one delay of at least one tunable buffer connected to the clock grid.